From: Paul Robinson <paul@...>
Date: 2006-08-31T22:47:40+09:00
Subject: Re: [ANN] How to spy on the Japanese Rubists

On 31 Aug 2006, at 14:25, Michal Suchanek wrote:

> Heh, it's not so.
>
> Have you ever tried to learn a foriegn language?

Only romance languages - French is particularly easy if you're  
natively English and know a little about construction of sentences  
from latin.

> For one, I heared that Eskimos use some tens of various words for
> different kinds of snow and ice. You get the idea.

They're called inuits, not eskimos - calling somebody an eskimo is  
like calling them a nigger, i.e. highly offensive - but I'm aware of  
what you mean by the snow/ice thing.

However, this is just social slang, all cultures have it, and there  
is more slang for those things that culture is obsessed by. Think how  
many different terms there are in western culture for genitals and  
having sex and getting drunk (yes, I know this is a sad statement on  
western culture) - it's exactly the same thing. A big enough  
dictionary takes care of it.

> Even in areas where the needs for precision weren't very different the
> slicing often happens at different places. So one word would be
> translated as different words in different contexts, often in both
> ways.

That is just basic grammar though - if you know that that a pronoun  
references an antecedent, and you can spot the antecedent, you can  
deal with the pronoun correctly. If you are able to identify the  
subject and predicate of a sentence and identify a passive or active  
voice, spot tense, etc. then you can translate easily.

This isn't hard - a 2-year old baby can do it with very primitive  
language skills.

My point is that current translation technology doesn't even attempt  
to try and do that.

> And the grammar is far from equivalent. I already found English
> sentences that I can understand (or so I think) but which would need
> several sentences to be explained in my native language.
> And these are both Indoeuropean languages. Japanese grammar is much
> more interesting :)

Maybe translation should move away from the literal word-for-word  
translation of words and move toward being able to express an  
identical idea or thought.

> The only correct translation can come from analyzing the meaning of a
> sentence in context of the previous text, and constructing sentence(s)
> in the other language with similar meaning. Of course, this is nearly
> impossible to do with a computer.

Shouldn't be - all a brain is, is a computer - and I know that gcc  
can parse syntax within a context far more accurately than I can.  
What you actually need is a compiler of natural languages - a very,  
very big syntax parsing mechanism - and then we have all the right  
tools we need. If somebody is able to write a tool that can translate  
ruby into C accurately, I don't see why somebody can't write a tool  
that can translate Japanese into English using a similar set of methods.

Anyway, enough of that, we're way off topic. Interesting discussion. :-)

--
Paul